Technical Field
The present invention generally relates to a waveguide architecture for a photonic neural component, and more particularly to a waveguide architecture for e.g., a photonic neural component of a neural network.
Related Art
Non-traditional, neuromorphic computing architectures such as neural networks and reservoir computing have shown promise in terms of performance, but conventional electronic approaches to interconnecting neurons have met with some limitations. For example, the IBM TrueNorth system operates with a processing speed in the kHz range due to the need for time multiplexing. Recently, excitable opto-electronics devices have generated interest as a way of potentially lifting this speed limitation. (See, for example, A. N., Tait et al., “Broadcast and Weight: An Integrated Network For Scalable Photonic Spike Processing,” J. Light. Tech. 32, 3427, 2014, M. A. Nahmias et al., “An integrated analog O/E/O link for multi-channel laser neurons,” Appl. Phys. Lett. 108, 151106 (2016), and K. Vandoorne et al., “Experimental demonstration of reservoir computing on a silicon photonics chip,” Nature Communication 5, 3541, 2014). However, such attempts have been limited by very high power consumption and optical loss. Meanwhile, the fabrication of waveguide crossing structures with very low loss has recently become possible. (See, for example, N. Bamiedakis et al., “Low Loss and Low Crosstalk Multimode Polymer Waveguide Crossings for High-Speed Optical Interconnects,” 2007 Conference on Lasers and Electro-Optics (CLEO), CMG1).